yeast: Support inherited leaf content in rules!

A bare leaf used as a rule result has an obvious value: the value of the
node matched by the rule. Using the output schema we can now distinguish
leaves from fieldless structural nodes and mark top-level bare leaves as
inherited content.

For nested bare leaves, there's no obvious input node from which to
inherit their value, so these still need to be handled explicitly.

Author: tausbn

Cargo.lock

@@ -10,3 +10,4 @@ proc-macro = true
 proc-macro2 = "1.0"
 quote = "1.0"
 syn = { version = "3.0", features = ["full"] }
+yeast-schema = { path = "../yeast-schema" }

shared/yeast-macros/Cargo.toml

@@ -1,3 +1,5 @@
+use std::collections::BTreeSet;
+
 use proc_macro2::{Ident, Literal};
 
 /// A complete rewrite rule such as `(input) where guard => (output)`.
@@ -117,6 +119,8 @@ pub(crate) enum Content {
     Static(Literal),
     /// Computed leaf content such as `#{name}` in `(identifier #{name})`.
     Computed(syn::Block),
+    /// Value inherited from the node matched by the enclosing rule.
+    Inherited,
 }
 
 pub(crate) enum Field {
@@ -137,3 +141,117 @@ pub(crate) enum Field {
         block: syn::Block,
     },
 }
+
+impl Rule {
+    pub(crate) fn resolve_bare_leaf_content(
+        &mut self,
+        leaf_kinds: &BTreeSet<String>,
+    ) -> syn::Result<()> {
+        let Replacement::Templates(templates) = &mut self.replacement else {
+            return Ok(());
+        };
+        for template in templates {
+            template.resolve_bare_leaf_content(leaf_kinds, true)?;
+        }
+        Ok(())
+    }
+}
+
+impl Template {
+    fn resolve_bare_leaf_content(
+        &mut self,
+        leaf_kinds: &BTreeSet<String>,
+        is_rule_result: bool,
+    ) -> syn::Result<()> {
+        let Template::Node(node) = self else {
+            return Ok(());
+        };
+
+        for field in &mut node.fields {
+            if let Field::Named { value, .. } = field {
+                value.resolve_bare_leaf_content(leaf_kinds, false)?;
+            }
+        }
+
+        if node.content.is_none()
+            && node.fields.is_empty()
+            && leaf_kinds.contains(node.kind.to_string().as_str())
+        {
+            if !is_rule_result {
+                return Err(syn::Error::new_spanned(
+                    &node.kind,
+                    format!(
+                        "bare leaf `{}` must be a top-level rule result or have explicit content",
+                        node.kind
+                    ),
+                ));
+            }
+            node.content = Some(Content::Inherited);
+        }
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use quote::quote;
+
+    fn output_leaf_kinds() -> BTreeSet<String> {
+        yeast_schema::node_types_yaml::leaf_kinds_from_yaml(
+            r#"
+named:
+  assignment:
+    left: identifier
+  identifier:
+"#,
+        )
+        .unwrap()
+    }
+
+    #[test]
+    fn marks_top_level_bare_leaves_as_inherited() {
+        let mut rule = crate::rule_parse::parse_rule(quote!((source) => (identifier))).unwrap();
+        rule.resolve_bare_leaf_content(&output_leaf_kinds())
+            .unwrap();
+
+        let Replacement::Templates(templates) = rule.replacement else {
+            panic!("expected template replacement");
+        };
+        let [Template::Node(node)] = templates.as_slice() else {
+            panic!("expected one output node");
+        };
+        assert!(matches!(node.content, Some(Content::Inherited)));
+    }
+
+    #[test]
+    fn rejects_nested_bare_leaves() {
+        let mut rule =
+            crate::rule_parse::parse_rule(quote!((source) => (assignment left: (identifier))))
+                .unwrap();
+        let error = rule
+            .resolve_bare_leaf_content(&output_leaf_kinds())
+            .unwrap_err();
+
+        assert_eq!(
+            error.to_string(),
+            "bare leaf `identifier` must be a top-level rule result or have explicit content"
+        );
+    }
+
+    #[test]
+    fn leaves_fieldless_structural_nodes_unchanged() {
+        let mut rule = crate::rule_parse::parse_rule(quote!((source) => (assignment))).unwrap();
+        rule.resolve_bare_leaf_content(&output_leaf_kinds())
+            .unwrap();
+
+        let Replacement::Templates(templates) = rule.replacement else {
+            panic!("expected template replacement");
+        };
+        let [Template::Node(node)] = templates.as_slice() else {
+            panic!("expected one output node");
+        };
+        assert!(node.content.is_none());
+    }
+}

shared/yeast-macros/src/ast.rs

@@ -201,8 +201,18 @@ pub fn rule(input: TokenStream) -> TokenStream {
 /// Paths are resolved relative to the consuming crate's `CARGO_MANIFEST_DIR`
 /// (the same convention `include_str!` uses for relative paths). The
 /// resolved paths are also emitted as `include_str!` references so the
-/// consuming crate gets invalidated when a schema YAML changes, prepping
-/// the ground for compile-time type-checking against those schemas.
+/// consuming crate gets invalidated when a schema YAML changes.
+///
+/// A bare output leaf used as a top-level rule result inherits the matched
+/// input node's value:
+///
+/// ```text
+/// rule!((superExpr) => (super_expr))
+/// ```
+///
+/// Whether the output kind is a leaf is determined from the output schema.
+/// Nested bare leaves are rejected at compile time because they have no
+/// unambiguous input node from which to inherit a value.
 #[proc_macro]
 pub fn rules(input: TokenStream) -> TokenStream {
     let input2: TokenStream2 = input.into();

shared/yeast-macros/src/lib.rs

@@ -380,6 +380,7 @@ impl Node {
                         }
                     }
                 }
+                Content::Inherited => quote! { #ctx.inherited_literal(#kind_str) },
             };
             quote! { let __node: yeast::OutputId = #node; }
         } else {

shared/yeast-macros/src/lower.rs

@@ -51,6 +51,25 @@ pub fn parse_rules_top(input: TokenStream) -> Result<TokenStream> {
     };
     let input_abs = resolve_path(&input_path.value);
     let output_abs = resolve_path(&output_path.value);
+    let output_yaml = std::fs::read_to_string(&output_abs).map_err(|error| {
+        syn::Error::new(
+            output_path.literal_span,
+            format!(
+                "rules!: failed to read output schema `{}`: {error}",
+                output_abs.display()
+            ),
+        )
+    })?;
+    let output_leaf_kinds = yeast_schema::node_types_yaml::leaf_kinds_from_yaml(&output_yaml)
+        .map_err(|error| {
+            syn::Error::new(
+                output_path.literal_span,
+                format!(
+                    "rules!: failed to parse output schema `{}`: {error}",
+                    output_abs.display()
+                ),
+            )
+        })?;
 
     let list = expect_group(&mut tokens, proc_macro2::Delimiter::Bracket)?;
     if let Some(token) = tokens.next() {
@@ -69,7 +88,9 @@ pub fn parse_rules_top(input: TokenStream) -> Result<TokenStream> {
                 explicit_rule_body(&item)
             };
             if let Some(rule_body) = rule_body {
-                Ok(crate::rule_parse::parse_rule(rule_body)?.lower())
+                let mut rule = crate::rule_parse::parse_rule(rule_body)?;
+                rule.resolve_bare_leaf_content(&output_leaf_kinds)?;
+                Ok(rule.lower())
             } else {
                 Ok(item)
             }
@@ -89,6 +110,7 @@ pub fn parse_rules_top(input: TokenStream) -> Result<TokenStream> {
 
 struct NamedString {
     value: String,
+    literal_span: Span,
 }
 
 fn parse_named_string_arg(
@@ -114,7 +136,10 @@ fn parse_named_string_arg(
             format!("`{expected_name}` must be a string literal path"),
         )
     })?;
-    Ok(NamedString { value })
+    Ok(NamedString {
+        value,
+        literal_span: literal.span(),
+    })
 }
 
 fn string_literal_value(literal: &proc_macro2::Literal) -> Option<String> {

shared/yeast-macros/src/parse.rs

@@ -236,6 +236,27 @@ impl<'a, C> BuildCtx<'a, C> {
         )
     }
 
+    /// Create a leaf whose value is inherited from the current rule's match.
+    #[doc(hidden)]
+    pub fn inherited_literal(&mut self, kind: &'static str) -> OutputId {
+        let source = self
+            .translator
+            .as_ref()
+            .expect("inherited leaf content requires a rule match")
+            .matched_root();
+        let node = self
+            .ast
+            .get_node(source)
+            .expect("rule source node must exist");
+        let stored = node.opt_string_content();
+        let has_children = node.fields.values().any(|children| !children.is_empty());
+        let content = match stored {
+            Some(value) if !value.is_empty() || !has_children => value,
+            _ => self.ast.source_text(source),
+        };
+        self.create_named_token_with_range(kind, content, None)
+    }
+
     /// Create a leaf node with a fixed string content.
     pub fn literal(&mut self, kind: &'static str, value: &str) -> OutputId {
         self.create_named_token_with_range(kind, value.to_string(), None)

shared/yeast/src/build.rs

@@ -1117,6 +1117,12 @@ impl<C> Clone for TranslatorHandle<'_, C> {
     }
 }
 
+impl<C> TranslatorHandle<'_, C> {
+    pub(crate) fn matched_root(&self) -> Id {
+        self.matched_root
+    }
+}
+
 impl<'a, C: Clone> TranslatorHandle<'a, C> {
     /// Recursively apply the current phase's rules to `id` and return the
     /// resulting node ids.

shared/yeast/src/lib.rs

@@ -109,6 +109,21 @@ fn run_and_ast(input: &str, rules: Vec<Rule>) -> OutputAst {
     runner.run(input).unwrap()
 }
 
+fn run_ast_with_rules(ast: Ast, rules: Vec<Rule>) -> OutputAst {
+    let schema = yeast::node_types_yaml::schema_from_yaml(OUTPUT_SCHEMA_YAML).unwrap();
+    let phases = [Phase::new("test", rules)];
+    let runner: Runner = Runner::with_schema_no_language(&schema, &phases);
+    runner.run_from_ast(ast).unwrap()
+}
+
+fn synthetic_integer_ast(content: &str) -> Ast {
+    let lang: tree_sitter::Language = tree_sitter_ruby::LANGUAGE.into();
+    let mut ast = Ast::with_schema(yeast::schema::from_language(&lang));
+    let root = ast.create_named_token("integer", content.to_owned());
+    ast.set_root(root);
+    ast
+}
+
 /// Helper: translate Ruby source with multiple rule phases and return its dump.
 fn run_phased_and_dump(input: &str, phases: Vec<Phase>) -> String {
     let lang: tree_sitter::Language = tree_sitter_ruby::LANGUAGE.into();
@@ -681,6 +696,74 @@ fn test_forward_scan_preserves_order() {
 
 // ---- Tree builder tests ----
 
+#[test]
+fn test_bare_leaf_rule_result_inherits_parsed_source_text() {
+    let rules: Vec<Rule> = yeast::rules! {
+        input: "tests/input-types.yml",
+        output: "tests/node-types.yml",
+        [
+            (integer) => (identifier),
+        ]
+    };
+    let ast = run_and_ast("x = 1", rules);
+    let dump = dump_ast(&ast, ast.get_root(), "x = 1");
+
+    assert_dump_eq(
+        &dump,
+        r#"
+        program
+          assignment
+            left: identifier "x"
+            right: identifier "1"
+    "#,
+    );
+    assert!(ast.reachable_node_ids().into_iter().any(|id| {
+        let node = ast.get_node(id).unwrap();
+        node.kind_name() == "identifier" && node.opt_string_content().as_deref() == Some("1")
+    }));
+}
+
+#[test]
+fn test_bare_leaf_rule_result_inherits_synthetic_content() {
+    let ast = run_ast_with_rules(
+        synthetic_integer_ast("synthetic"),
+        yeast::rules! {
+            input: "tests/input-types.yml",
+            output: "tests/node-types.yml",
+            [
+                rule!((integer) => (identifier)),
+            ]
+        },
+    );
+    let dump = dump_ast(&ast, ast.get_root(), "");
+
+    assert_dump_eq(
+        &dump,
+        r#"
+        identifier "synthetic"
+    "#,
+    );
+}
+
+#[test]
+fn test_bare_leaf_rule_result_inherits_empty_content() {
+    let ast = run_ast_with_rules(
+        synthetic_integer_ast(""),
+        yeast::rules! {
+            input: "tests/input-types.yml",
+            output: "tests/node-types.yml",
+            [
+                (integer) => (identifier),
+            ]
+        },
+    );
+
+    assert!(ast.reachable_node_ids().into_iter().any(|id| {
+        let node = ast.get_node(id).unwrap();
+        node.kind_name() == "identifier" && node.opt_string_content().as_deref() == Some("")
+    }));
+}
+
 #[test]
 fn test_tree_builder() {
     let input = "x = 1";